Search papers, labs, and topics across Lattice.
4
0
8
18
LLM safety is a cat-and-mouse game: ORPO excels at breaking alignment, while DPO is best at restoring it, but at the cost of overall usefulness.
LLM-based multi-agent systems can more than double their performance and slash token usage by organizing themselves like a company, with distinct governance, execution, and compliance layers.
Even the most advanced LLMs like GPT-5.2 and Gemini-3-Pro often fail to recognize and refuse to process harmful content embedded within seemingly harmless tasks.
Shadow APIs promising access to top LLMs like GPT-5 and Gemini 2.5 often deliver significantly degraded performance (down to 47.21% accuracy) and fail identity verification, casting doubt on research relying on them.
